Protect Privacy in a Distributed Learning Platform with a Natural Language Processing Example

Along with the development of Internet of Things, smart home, smart community, smart city,
intelligent transportation, and smart healthcare also are becoming novel paradigms that are rapidly
gaining widespread impact. Although the definition of “Things” changes as technologies evolve, the core
idea of the objects interacting with each other and cooperating with human-in-the-loop order to create
understandable knowledge, infer environmental context, and reach common goals remain unchanged.
Vast stores of digital data about environmental contexts and user behaviors are the basis of keen scientific
interest about what new insights can be gained by combining and 'mining' these rich, multi-dimensional,
heterogeneous data resources together. In such a distributed and mobile environment, it is widely
accepted that most of data generated by the massive number of IoT devices should be processed locally at
the devices or at the edge ends; otherwise, the total amount of data for a centralized cloud would
overwhelm the communication network bandwidth and incur potential security threats. In addition, many
data have strict privacy, security, and regulatory constraints, such that a centralized data collection
approach may not be applicable. Improper or non-existent disclosure control may violate these constraints
and cause personal information leakage. The ultimate goal for this proposal is to construct a
distributed learning platform on mobile devices as a service (PaaS) model, where Machine
Learning as a Service (MLaaS) are offered in a common network communication architecture and
training and deploying deep learning models, and where Database as a Service (DaaS) provides
user access and data privacy control. This proposed study will accelerate research on IoT applications
using mobile devices as a hub, distributed learning system software design, database security and privacy,
providing a standardized environment where data science experiments can be validated and compared.